[ML] Inference API chunking large documents

[jonathan-buttner]

Description

Large documents need to be chunked otherwise tokens exceeding the model's limit won't be used.

MVP for default word based chunking strategy:

• Use a sliding window approach
• Chunk into 200 words
• Try splitting on whitespace or newline if possible
• Maybe fallback to only doing it based on character length (don't have to search for whitespace/newlines, and also avoids problems with languages that don't really use whitespace)

MVP for configurable chunking settings:

• Allow users to provide chunking configurations to select between word and sentence based chunking strategies when creating an inference endpoint.
• Fallback to default word based chunking strategy above

Post MVP features for configurable chunking settings:

• Allow users to enable chunking when calling to perform an inference through the Inference API
• Allow users to configure chunking as part of their ingestion pipeline

Tasks already completed:

• [X] Create default word based chunking strategy https://github.com/elastic/elasticsearch/pull/107303
• [X] Add chunking configuration to all services
• [X] Update default strategy to sentence based chunking strategy
• [X] Add chunk size upper and lower limits
• [x] Remove feature flag

[elasticsearchmachine]

Pinging @elastic/ml-core (Team:ML)